Inadequacies of minimum spanning trees in molecular epidemiology

Minimum spanning trees (MSTs) are frequently used in molecular epidemiology research to estimate relationships among individual strains or isolates. Nevertheless, there are significant caveats to MST algorithms that have been largely ignored in molecular epidemiology studies and that have the potential to confound or alter the interpretation of the results of those analyses. Specifically, (i) presenting a single, arbitrarily selected MST illustrates only one of potentially many equally optimal solutions, and (ii) statistical metrics are not used to assess the credibility of MST estimations. Here, we survey published MSTs previously used to infer microbial population structure in order to determine the effect of these factors. We propose a technique to estimate the number of alternative MSTs for a data set and find that multiple MSTs exist for each case in our survey. By implementing a bootstrapping metric to evaluate the reliability of alternative MST solutions, we discover that they encompass a wide range of credibility values. On the basis of these observations, we conclude that current approaches to studying population structure using MSTs are inadequate. We instead propose a systematic approach to MST estimation that bases analyses on the optimal computation of an input distance matrix, provides information about the number and configurations of alternative MSTs, and allows identification of the most credible MST or MSTs by using a bootstrapping metric. It is our hope this algorithm will become the new "gold standard" approach for analyzing MSTs for molecular epidemiology so that this generally useful computational approach can be used informatively and to its full potential.